The rationing of medical care prioritizes the need for early predictors of death in the surgical intensive care unit (SICU). We prospectively studied 100 consecutive SICU admissions, looking for predictors of early death in the SICU and the cost implications of these findings. Serial APACHE II scores on days 1, 3, and 5 were subjected to multinomial logistic regression analysis to determine significant predictors of death in the SICU on day 1. Survivors had significantly lower (p less than 0.05) mean day-1 APACHE II scores than had nonsurvivors (13.6 vs 22.1). Half of the patients with scores greater than 18 died, and all patients with scores on day 1 of 25 or greater died. Significant predictors of death on SICU day 1 were APACHE II scores, Acute Physiology Score, Glasgow Coma Score, creatinine level, and Chronic Health Evaluation Score. Forty-one patients had been transferred from community hospitals as a results of acute illness; this population accounted for two thirds of the deaths in the SICU. Ten of 18 nonsurvivors were predicted on day 1, with these patients incurring a total cost of approximately $1 million. If therapy had been modified on days 5, 10, or 15, the potential cost savings would have been $340,000, $240,000, or $140,000, respectively. Integration of the results of this study into the management decision-making process and treatment guidelines may reduce the cost of care in the SICU.